The first line of defence is the business teams that own and manage risk day to day: they run processes, operate the controls and answer for the outcomes. In the three lines model, a second line oversees risk and a third gives independent assurance. When a team deploys AI agents, the agents and their review sit in the first line.
Why does this matter with agents?
Because deploying an agent does not move responsibility. A claims team that uses an agent to settle small claims still owns those settlements. A payments team that lets an agent draft payments still owns every payment that goes out. The temptation is to treat the agent as something technology or risk owns, which leaves nobody in the business accountable for what it does.
Keeping agents in the first line has a practical consequence for training. The people who review held actions, decide exceptions and bank corrections are first-line staff. Their competence is the control. If they cannot read the agent's verdicts or judge the provenance of a fact, the first line has handed its job to software without anyone noticing.
What does the first line do in an agent deployment?
- Owns the agent. A named agent owner in the business answers for its brief, sources, bounds and certification.
- Reviews held actions. Certified reviewers decide what the agent may not complete alone.
- Operates the controls. Approval limits, red lines, rotas and response-time targets are run day to day by the team.
- Monitors the record. Team leads read patterns in verdicts and corrections, and act on them.
- Reports issues. Incidents and control failures go to the second line promptly.
An example
Take a hypothetical mortgage lender. An agent assembles application files and drafts affordability assessments, which underwriters decide. The underwriting team is the first line: its manager owns the agent, underwriters review and decide, and team leads run weekly calibration. The risk function, as the second line, sets the standard for what evidence underwriters must check and challenges whether the agent's depth is appropriate. Internal audit, the third line, later tests whether underwriters actually checked that evidence, by sampling the record.
How is it different from the second and third lines?
| Line | Who | Role with agents |
|---|---|---|
| First | Business teams | Own, run and review the agents in their process |
| Second | Risk, compliance | Set standards, challenge design, monitor |
| Third | Internal audit | Test independently that the system works |
The first line does; the second line oversees; the third line checks. The lines blur when the second line is asked to approve every agent decision, which turns oversight into operation and leaves nobody independent to challenge it.
Should the second line be involved in first-line training?
Yes, without taking it over. When the second line sits in on first-line agent training, both sides learn the same vocabulary and the same red lines, and challenge becomes better informed. The case is made in why the second line should sit in your AI agent training, with the method in how to involve the second line in AI agent training. The guide to AI agent skills by role maps first-line roles in detail.
Where the Academy fits
Pryme Intelligence Academy's Operator track is designed for first-line staff: reviewers, team leads and operations managers who run a live workspace every day. A business can also adopt a practice of making first-line approval rights depend on a current Operator certificate, depth by depth, so the record shows who was qualified to decide.
Questions people ask
Is the agent itself part of the first line?
The agent works inside the first line's process, but it is not accountable for anything. Accountability rests with the named people who own it and review its work.
Can a small business use the three lines model?
Yes, in a lighter form. The principle that those who run a process are not the only ones checking it applies at any size.